Xiaomi's MiMo v2.6 tops open-model index, RL run cost ~$3.5M and was livestreamed
Prompt Engineering · youtube · 2026-09-22
Xiaomi released MiMo v2.6, now the top open model on the Artificial Analysis index, and livestreamed the entire RL run behind it, which cost roughly $3.5 million.
Key points from the hands-on review:
- The Pro model was tested in a DeepSeek harness on agentic tasks with verification loops: a self-researched landing page, a live ISS tracker, Blender assets, and Three.js scenes
- The video breaks down the VRAM actually needed to run Pro and Flash locally, arguing "open" doesn't mean "runnable at home"
- A Qwen-9B-distilled variant (MiMo-V2.6-Distill-Qwen-9B) is on Hugging Face
More from Models
- GGUF models can now run directly in Hugging Face transformers with ggml Metal kernels — pcuenq · 2026-09-22
- As OpenAI Preps a ChatGPT Bot to Rival Meta Muse and Grok, Here's What Personal AI Assistants Can Already Do — Efistoffeles · 2026-09-22
- AI agent beats Pokemon in 50 hours; expert human speedrun pace (2 hours) seen within a year — ben_j_todd · 2026-09-22
- New Claude model beats Pokemon in 50 hours, 5x faster than Opus 4.7 five months ago — ben_j_todd · 2026-09-22
- Developer: Most Jev Use Cases Are Useless, but Browser Automation Is Where It Shines — EXM7777 · 2026-09-22
- ChatGPT Co-Inventor Launches Jev: Claims 20-200x Faster, 40-400x Cheaper Than LLMs — zmkzmkz · 2026-09-22